Dear Listers,
I have asked the list a question on the above subject on 24/September/03.
Here is the question and the list of answers I have received, with many
thanks to those who took the time to respond.
Salma
Question:
My question was asked in July 2000 by Maurizio as below. Is it possible to
acess the answers of this question? if not may I ask the same question
again!.
I very much appreciate your advice!
---------
Dear Bugs Listers, I usually analyze survey data with probability weights.
Is there any way to incorporate such weights in Bayesian analysis using
WinBugs? Thank you very much for any help and best wishes,
Maurizio Pisati
------------------
Answers:
(1)
You just add the weights as you write the likelihood function. For example,
if the probability weight of a normal distribution is p(1-p), you can use:
t[i] ~ dnorm(g[i], tau[i])
tau[i]<- p[i]/(g[i]*(1-g[i]))
Ernesto Calvo
(2.a)
In case it helps, there are some examples in Peter Congdon's first book,
Bayesian Statistical Modelling, on estimating population parameters from
survey data (see p. 190 - 194). Unfortunately, those examples don't involve
probability weights! They involve data generated from a "two-stage sampling
design", i.e. strata at stage one and then a random sample of individuals
within the selected PSU's ... (see the middle of p. 191).
Maybe a crude workaraound would be to simulate data based on the
probability weights from a survey, so as to obtain data akin to a two-stage
sampling design, and then apply the approach discussed by Congdon using the
simulated data ?? I might be way off here, so feel free to reign me in !!
Jagat Sheth
(2.b)
Salma,
Thanks alot. I just remembered about the standard errors for parameters in
a model fitted to complex survey data - they may need to be adjusted for
the clustering effects at PSU level? Wonder if WinBUGS code exists for
this, if its necessary.
Thanks again for posting any replies back on the list. Looking forward to
reading them as and when possible.
Best wishes,
Jagat
(3)
I feel sure there must be a way of tricking the likelihood
to have the right form, but I can't immediately think of one
(except in the normal case, where the precision can be adjusted).
It might be worth asking on the list - or have you already done that?
David
(4)
I have seen it done before, but only for a normal linear model. There is a
discussion in the BUGS listserv archives.
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?S1=bugs
Henry
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Salma Ayis, PhD.
MRC HSRC,Department of Social Medicine
University of Bristol
Canynage Hall
Whiteladies Road
Bristol BS8 2PR
UK
Tel: +44 (0)117 9287361
Fax: +44(0)117 9287236
e-mail: [log in to unmask]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-------------------------------------------------------------------
This list is for discussion of modelling issues and the BUGS software.
For help with crashes and error messages, first mail [log in to unmask]
To mail the BUGS list, mail to [log in to unmask]
Before mailing, please check the archive at www.jiscmail.ac.uk/lists/bugs.html
Please do not mail attachments to the list.
To leave the BUGS list, send LEAVE BUGS to [log in to unmask]
If this fails, mail [log in to unmask], NOT the whole list
|